How can I use the docred dataset with strict evaluation about rebel HOT 5 CLOSED

The output is like [{'head': 'Skai TV', 'type': 'headquarters location', 'tail': 'Piraeus'}].
How can I adjust the model to output (start_idx (inclusive), end_idx (exclusive))

from rebel.

Thank you very much for bringing this up.

I somehow lost two lines when cleaning the code for release. I have updated the code with the fix:

This will also extract the entity types and won't return an error.

Sorry for the inconvenience.

from rebel.

Thank you for this! I had just fixed it. And I want to ask what is the function of conf.finetune?
The default value is False. Does it mean that model will be trainned from scratch? Do I need to adjust it to True to finetune on my dataset?

from rebel.

Hi, that is some legacy parameter, not really used anymore. The model is always trained from a BART checkpoint, but that is defined by "model_name_or_path", "config_name" and "tokenizer_name". You can finetune from just pretrained BART or from the REBEL pre-trained on the silver data checkpoint.

from rebel.

Thanks！

from rebel.